energies

Article
Evaluation of a 1 MW, 250 kW-hr Battery Energy
Storage System for Grid Services for the Island
of Hawaii
Karl Stein 1 , Moe Tun 2 , Keith Musser 3 and Richard Rocheleau 4, *
1 Center for Climate Physics, Institute for Basic Science (IBS), Busan 46241, Korea; kjstein@gmail.com
2 HNU Photonics LLC, Kahului, HI 96732, USA; moetunhawaii@gmail.com
3 Integrated Dynamics, Inc., Fishers, IN 46037, USA; kmusser@idi-software.com
4 Hawai‘i Natural Energy Institute, SOEST, University of Hawaii at Mānoa, Honolulu, HI 96822, USA
* Correspondence: rochelea@hawaii.edu




Received: 27 October 2018; Accepted: 22 November 2018; Published: 1 December 2018


Abstract: Battery energy storage systems (BESSs) are being deployed on electrical grids in significant
numbers to provide fast-response services. These systems are normally procured by the end user,
such as a utility grid owner or independent power producer. This paper introduces a novel research
project in which a research institution has purchased a 1 MW BESS and turned ownership over to a
utility company under an agreement that allowed the institution to perform experimentation and
data collection on the grid for a multi-year period. This arrangement, along with protocols governing
experimentation, has created a unique research opportunity to actively and systematically test the
impact of a BESS on a live island grid. The 2012 installation and commissioning of the BESS was
facilitated by a partnership between the Hawaii Natural Energy Institute (HNEI) and the utility
owner, the Hawaiian Electric and Light Company (HELCO). After the test period ended, HELCO
continued to allow data collection (including health testing). In 2018, after 8500 equivalent cycles, the
BESS continues to operate within specifications. HNEI continues to provide HELCO with expertise
to aid with diagnostics as needed. Details about the BESS design, installation, experimental protocols,
initial results, and lessons learned are presented in this paper.

Keywords: battery energy storage system; field evaluation; wind smoothing; frequency regulation;
grid-scale; lithium-titanate

1. Introduction
At high penetration levels, the integration of intermittent renewable energy sources poses several
challenges for grid operations, due, in part, to the variability of renewable energy sources, but also
due to the reduction of system inertia via the displacement of traditional dispatchable generation.
These factors can result in a reduction in grid stability and reliability, manifesting in such ways as
increased frequency variability, voltage transients, and power quality reduction [1–3]. Such effects are
magnified on small island grids such as those on the Hawaiian islands [4]. The geographic isolation of
the islands’ electricity grids and the rapid growth of renewable generation make the Hawaiian grids
particularly susceptible to the adverse effects of variable renewable energy sources, but also an ideal
test-bed for various emerging grid solutions, including energy storage. With already high penetration
of wind and solar generation, recently passed legislation set a goal of 100% renewable energy by the
year 2045 for the state [5] and recent utility planning ensures that the islands’ grids will see increasing
renewable energy penetration in the coming years. Similar challenges and solutions are also being
addressed on other island grids [6–10].

Energies 2018, 11, 3367; doi:10.3390/en11123367 www.mdpi.com/journal/energies

Energies 2018, 11, 3367 2 of 17

Fast-acting battery energy storage systems (BESSs) show promise in mitigating many of the effects
of high renewable energy penetration levels [11–15]. Despite substantial numbers of deployments
worldwide, few studies have reported results of grid-connected systems [16–19]. This paper presents
a unique field test of a BESS operating on an island electric grid which was made possible by a
partnership between the Hawaii Natural Energy Institute (HNEI), at the University of Hawaii at
Mānoa, and the utility grid operator, the Hawaii Electric Light Company (HELCO, Hilo, HI, USA).
In accordance with the agreement, HNEI purchased the BESS, then relinquished ownership to HELCO,
but was allowed to perform experiments with the BESS while it was operating on the electric grid.
The experiments involved switching the BESS on and off every 20 min. This switching forms both an
independent and dependent variable over a relatively short time (minimizing changes in background
conditions). This allowed for much more direct measurements of the effect of the BESS on the island
grid than would be possible with data from long-term monitoring of the grid before and after BESS
installation. To the best of our knowledge, no other BESS research projects involve a multi-year
agreement that allows for robust scientific experimentation.
The work presented here is part of a larger research program at HNEI, which includes assessment
of battery performance on the grid, optimization of control algorithms to maximize grid support while
minimizing battery cycling [20], and field and laboratory testing of cells to better understand cell
aging and degradation [21,22]. This paper focuses on the grid support performance aspect, along with
initial testing and methodology. The paper describes Hawaii Island BESS project including project
development and installation (Section 2), development of the control algorithms, acceptance test
results, and initial performance tests (Section 3) followed by concluding remarks and the plans for
future work (Section 4).

2. BESS Development and Installation

The investigation into use of a BESS system for the island of Hawaii was initiated as a result
of a 2009 analysis of that electrical grid performed by General Electric (GE, Schenectady, NY, USA)
under contract to HNEI. The study utilized the Positive Sequence Load Flow (PSLF) power system
analysis software and historical data to model the frequency response of the Hawaii island grid to
rapid changes in wind generation and other contingency events. At the time of the study, the electric
grid, operated by the Hawaii Electric Light Company (HELCO), had a peak load of approximately
180 MW and a wind capacity of approximately 32 MW. While there has been rapid growth since, there
was negligible photovoltaics on the system at that time. The models indicated that as little as 1 MW
of short-duration fast-acting energy could significantly reduce the severity and duration of the grid
frequency events.
These results prompted HNEI to develop a research project to procure and evaluate a 1 MW
grid-connected BESS. HNEI and the grid owner, HELCO, reached a mutually beneficial memorandum
of understanding that would facilitate the realization of this research project (and address any liability
issue). In accordance with the agreement, HNEI procured the BESS, and transferred ownership over
to HELCO immediately after commissioning. The agreement further stipulated that HELCO would
provide HNEI with the ability to perform experiments on the grid (provided adherence to protocol
that was agreed upon by both parties). A lithium ion titanate battery chemistry was chosen due to
the desire for an extended cycling lifetime and faster charge/discharge rates compared to the more
common carbon anode electrochemistry [23].
HNEI assembled a public-private partnership (Table 1) for BESS development, installation,
and testing. The key tasks included: site preparation, material procurement, electrical design,
communications design, and algorithm development. Procurement of balance of plant (BOP) materials
(e.g., transformers, meters, and compatible communications devices) was undertaken by HELCO,
Altairnano (Anderson, IN, USA), and HNEI. The site preparation was conducted jointly by Altairnano
and the installation site land owner, Haw’i Renewable Development (HRD, Upolu Point, HI, USA),
which owns a collocated 10.6 MW wind farm. The electrical and communications systems were
Energies
Energies 2018,
2018, 11,
11, x FOR PEER REVIEW
3367 33of
of 17
17

Dynamics Inc. (IDI, Fishers, IN, USA) and SCADA Solutions Inc. (Irvine, CA, USA). HNEI provided
designed management
program by HELCO and Altairnano.
and Algorithmindevelopment
technical oversight wasdevelopment
all areas of the performed byeffort.
Integrated Dynamics
Inc. (IDI,
The BESS was installed at a transmission level site on the grid, at the point of provided
Fishers, IN, USA) and SCADA Solutions Inc. (Irvine, CA, USA). HNEI program
common coupling
management
(PCC) between and thetechnical
HRD wind oversight
farm andin all
theareas of the
nearby development
Waimea effort.
substation. Site construction started in
The BESS was installed at a transmission level site
April of 2012 and the BESS was commissioned in December of 2012. Theon the grid, at the point
BESSof common coupling
was designed for
(PCC) between the HRD wind farm and the nearby Waimea substation. Site
interconnection to electrical power systems (as defined by ANSI C84.12006), and to be compliant construction started
with
in April ofstandards
applicable 2012 and the BESS
during was commissioned
discharge in December
(IEEE 15472003) of 2012. The BESS
and standby/charging modes was designed
(IEEE for
519) [24].
interconnection to electrical power systems (as defined by ANSI C84.12006), and to be compliant with
applicable
Table 1.standards during
List of funding discharge
sources (IEEEfor
and partners 15472003) and standby/charging
Hawaii Battery modes
Energy Storage Systems (IEEE
(BESS) 519) [24].
project.

Project
Table 1. List of funding sources and partners for Hawaii Management
Battery Energy Storage Systems (BESS) project.
Hawaii Natural Energy Institute (HNEI) Algorithm Development
Technical Oversight
Project Management
Hawaii Natural Energy Institute (HNEI) Algorithm
Primary Development
Funding Source
Office of Naval Research (ONR) Technical Oversight
Program Oversight
Primary Funding Source
Office ofof
Department Naval Research
Energy (DOE) (ONR) Partial Funding for Algorithm Development
Program Oversight
Department of Energy (DOE) Infrastructure Development
Partial Funding for Algorithm Development
Hawaii Electric Light Company (HELCO) Planning
Infrastructure Development
Hawaii Electric Light Company (HELCO) Planning
Grid Management
Grid Management
Host Site Owner
Haw’i Renewable Development (HRD) Host Site Owner
Haw’i Renewable Development (HRD) Site Preparation
Site Preparation
Battery
BatterySystem
SystemManufacturer
Manufacturer
Altairnano
Altairnano Systems Integration
Systems Integration
FrequencyAlgorithm
Frequency AlgorithmDevelopment
Development
Integrated
Integrated Dynamics,
Dynamics, Inc. (IDI)
Inc. (IDI) SoftwareDevelopment
Software Development
SCADA Solutions Wind Algorithm Development
SCADA Solutions Wind Algorithm Development
Parker–Hannifin Company Power Conversion System Supplier
Parker–Hannifin Company Power Conversion System Supplier

The system
The system isis housed
housed in
in two
two containers,
containers, one
one for
for the
the power
power module
module (PM,
(PM, Figure
Figure 1a)
1a) and
and aa second
second
for the
for the power
power conversion
conversion system
system(PCS,
(PCS,Figure
Figure1b).
1b).The
ThePMPMcontains
containsthe
thebattery
batterystack,
stack,along
along with
with a
a ground fault detection system, a fire suppression system, HVAC temperature control
ground fault detection system, a fire suppression system, HVAC temperature control system, the system, the
battery management
battery management system
system(BMS),
(BMS),andandthe
thesite
sitedispatch
dispatchcontroller
controller(SDC).
(SDC).The BMS
The BMSis responsible for
is responsible
monitoring,
for controlling,
monitoring, and and
controlling, protecting the battery
protecting cells, including
the battery monitoring
cells, including state-of-charge
monitoring (SOC),
state-of-charge
(SOC), preventing over-charge/under-charge, and protecting against thermal damage. The PCSa
preventing over-charge/under-charge, and protecting against thermal damage. The PCS contains
four-quadrant
contains Parker-Hannifin
a four-quadrant inverter (+/−inverter
Parker-Hannifin real and(+/−
reactive
real power), cooling
and reactive system,cooling
power), metering units,
system,
processing units, and associated protection.
metering units, processing units, and associated protection.

(a) (b)
Figure
Figure 1.
1. Containers
Containers for
for the
the BESS
BESS (a)
(a) power
power module
module (PM);
(PM); and
and (b)
(b) power
power conversion
conversion system
system (PCS).
(PCS).
Energies 2018, 11, 3367 4 of 17

The control center of the BESS is the site-dispatch controller housed in the power module, which
performs the
Energies following
2018, tasks:
11, x FOR PEER REVIEW 4 of 17

1. Collects and stores

The control allofdata
center on the
the BESS cloud
is the server controller housed in the power module, which
site-dispatch
2. Utilizes data
performs from thetasks:
the following BMS (e.g., SOC and cell temperatures) and the inverter system in the PCS
to determine
1. any
Collects and limits
stores allon available
data power
on the cloud server
3. Processes the measured data to develop
2. Utilizes data from the BMS (e.g., SOC and power commands using
cell temperatures) and thereal-time controlinalgorithms
inverter system the PCS
4. to determine
Executes manualany limits on available
commands inputtedpower
by a user
3. Processes the measured data to develop power commands using real-time control algorithms
The SDC coordinates
4. Executes the PM and
manual commands PCSby
inputted toaprovide
user the desired amount of real and/or reactive
power at the point of common connection (PCC). It can accept automatic generation control (While the
The SDC coordinates the PM and PCS to provide the desired amount of real and/or reactive
AGCpowercapability exists, it was never connected or utilized.) (AGC) commands or operate in manual or
at the point of common connection (PCC). It can accept automatic generation control (While
automatic modes.
the AGC capability In manual
exists, itmode, the connected
was never operator specifies the(AGC)
or utilized.) amount of real/reactive
commands or operatepower desired,
in manual
whileorinautomatic
automaticmodes. mode In themanual
powermode,
command is calculated
the operator according
specifies the amountone ofof two control algorithms
real/reactive power
(Section 3). Operators
desired, interact mode
while in automatic with the
the SDC
powervia a PC-based
command Humanaccording
is calculated Machineone Interface
of two (HMI)
control over
a secure network
algorithms connection.
(Section Theseinteract
3). Operators interactions
with the typically
SDC via include:
a PC-based selection
Human of controlInterface
Machine algorithms,
(HMI)parameter
algorithm over a secure network
settings, connection.orThese
diagnostics, interactions
inputting of manual typically include: Manual
commands. selection commands
of control are
oftenalgorithms,
used to executealgorithm parameter settings,
State-of-Health diagnostics,
tests which or inputting
are performed of manualscheduled
at regularly commands.intervals.
Manual
commands are often used to execute State-of-Health tests which are performed at regularly
Figure 2 shows a simplified diagram of the system, indicating the location of the four meters
scheduled intervals.
used for data collection. There are three Schweitzer SEL-735 m which sample current, voltage, and
Figure 2 shows a simplified diagram of the system, indicating the location of the four meters
frequency; one each for the BESS (“Battery”), the wind farm (“Wind Farm”), and combined signals
used for data collection. There are three Schweitzer SEL-735 m which sample current, voltage, and
(“Totalizer”).
frequency; There is also
one each forathe
Shark
BESS200 m that can
(“Battery”), thebe usedfarm
wind to determine
(“Wind Farm”),the amount of auxiliary
and combined signalspower
that the
(“Totalizer”). There is also a Shark 200 m that can be used to determine the amount of auxiliary powerto the
BESS consumes. The meters sample at 10 Hz for control purposes. Data is recorded
cloudthatserver at a rate
the BESS of 5 HzThe
consumes. for meters
analysis. Within
sample theHz
at 10 PM,forbattery
control cells are organized
purposes. as follows:
Data is recorded to the7 cells
cloudgroup”,
to a “cell server at a rate
2 cell of 5 Hz
groups to for analysis.and
a module, Within
fourthe PM, battery
modules cellsreplaceable
to a line are organizedunitas(LRU,
follows: 7
Figure 2,
inset).cells to a “cell
Voltages forgroup”, 2 cell groups
each group (7 cells)toare
a module,
sampled andatfour
10 s modules
intervals, to and
a linetemperatures
replaceable unitfor(LRU,
groups of
Figure
28 cells 2, inset). consisting
(2 modules Voltages forofeach group
2 cell groups(7 cells)
each)are sampled
are recorded at 10 s intervals,
every minute. and temperatures
These for with
data, along
groups of 28 cells (2 modules consisting of 2 cell groups each) are recorded every minute. These data,
diagnostics and environmental data within the container, are stored in a cloud server for later analysis.
along with diagnostics and environmental data within the container, are stored in a cloud server for
7 cells are connected in parallel. 384 of these parallel groups are connected in series.
later analysis. 7 cells are connected in parallel. 384 of these parallel groups are connected in series.

Figure
Figure 2. Schematic
2. Schematic of the
of the Hawaiiisland
Hawaii islandBESS
BESS highlighting
highlighting metering
meteringunits. TheThe
units. pictures insets
pictures showshow
insets
the installed PM and PCS units on site (top), and the line replaceable units (LRUs) which
the installed PM and PCS units on site (top), and the line replaceable units (LRUs) which house house the the
battery cells within the PM (below).
battery cells within the PM (below).
Energies 2018, 11, 3367 5 of 17
Energies 2018, 11, x FOR PEER REVIEW 5 of 17

3. Control Algorithms
3. Control Algorithms
Most of the time, the Hawaii Island BESS operates in the automatic mode. In this mode, one
Most of the time, the Hawaii Island BESS operates in the automatic mode. In this mode, one of
of two algorithms can be selected, a wind smoothing algorithm or a frequency response algorithm.
two algorithms can be selected, a wind smoothing algorithm or a frequency response algorithm. The
The algorithms were developed by SCADA Solutions, LLC and Integrated Dynamics, Inc. (IDI) on
algorithms were developed by SCADA Solutions, LLC and Integrated Dynamics, Inc. (IDI) on behalf
behalf of the battery manufacturer, Altairnano. These algorithms were developed in close consultation
of the battery manufacturer, Altairnano. These algorithms were developed in close consultation with
with the utility (HELCO) and the HRD wind farm. Both algorithms were developed through a similar
the utility (HELCO) and the HRD wind farm. Both algorithms were developed through a similar
process required by HNEI (Figure 3) which is described below. Process ensures that the algorithms
process required by HNEI (Figure 3) which is described below. Process ensures that the algorithms
were meeting the design and performance objectives.
were meeting the design and performance objectives.
The development process began by defining a set of requirements and specifications for each
The development process began by defining a set of requirements and specifications for each
algorithm. Once these were agreed upon by all stakeholders, Altairnano and IDI designed the
algorithm. Once these were agreed upon by all stakeholders, Altairnano and IDI designed the
algorithms and HMI user interface to meet the specifications. Next, simple modeling studies were
algorithms and HMI user interface to meet the specifications. Next, simple modeling studies were
performed to optimize control algorithm parameters and to ensure the predicted behavior of the
performed to optimize control algorithm parameters and to ensure the predicted behavior of the BESS
BESS under each algorithm met the design objectives. After the algorithms passed an acceptance
under each algorithm met the design objectives. After the algorithms passed an acceptance test based
test based on the modeling studies, laboratory testing with computational hardware in-the-loop
on the modeling studies, laboratory testing with computational hardware in-the-loop was executed.
was executed. This step determined whether the algorithms, running on the embedded computers
This step determined whether the algorithms, running on the embedded computers performed as
performed as intended. Lastly, a site acceptance test (SAT) plan was developed and executed during
intended. Lastly, a site acceptance test (SAT) plan was developed and executed during BESS
BESS commissioning
commissioning to ensure
to ensure thatalgorithms
that the the algorithms performed
performed as intended
as intended in field.
in the the field. Descriptions
Descriptions of
of the
the two
two algorithms
algorithms andandthethe results
results of of
sitesite acceptance
acceptance andinitial
and initialtesting
testingofofthe
thealgorithms
algorithmsareare presented
presented
in the following sections.
in the following sections.

Figure
Figure 3.
3. Flowchart
Flowchart of
of the
the development
development process
process for
for the
the frequency
frequency response control algorithm.
response control algorithm.

3.1. Frequency
FrequencyResponse:
Response:Algorithm
AlgorithmDevelopment
Developmentand
andTesting
Testing
Some of the most important results in this section were made possible by the
the agreement
agreement between
HNEI and HELCO. As the section progresses, it will become apparent that quantifying the impact of
a frequency response algorithm would be limited without the unique ability of experimentalists to
actively engage in control of the BESS, and hence, the response of the grid. Unlike power smoothing
(including wind smoothing), it is not possible to directly determine how the grid frequency ‘would
have behaved’ if the BESS (running a frequency
frequency response
response algorithm)
algorithm) was not present.
present. This would
otherwise complicate the assessment of performance.
The electric grid on the island of Hawaii operates with a relatively high penetration of non-
dispatchable renewable resources (70 MW of wind and solar on a 180 MW grid in 2013), which can
result in increased frequency fluctuations due to mismatch of load and generation and reduced
Energies 2018, 11, 3367 6 of 17

The electric grid on the island of Hawaii operates with a relatively high penetration of
non-dispatchable renewable resources (70 MW of wind and solar on a 180 MW grid in 2013), which can
Energies 2018, 11, x FOR PEER REVIEW 6 of 17
result in increased frequency fluctuations due to mismatch of load and generation and reduced
rotational inertia.
rotational inertia. The The frequency
frequency response
response algorithm
algorithm waswas designed
designed toto mitigate
mitigate such
such frequency
frequency
deviations by sourcing and sinking real power to/from the
deviations by sourcing and sinking real power to/from the grid as needed. grid as needed.
The grid
The grid frequency
frequency isismeasured
measuredby bytwo
twometers:
meters:(a)(a)Shark
Shark200 m,m,
200 and
and(b)(b)
SEL-735 m (Figure
SEL-735 m (Figure 2, top
2,
left of Figure 4). The Shark meter proved to have a significant lag due to filtering, so
top left of Figure 4). The Shark meter proved to have a significant lag due to filtering, so the frequencythe frequency
signal is
signal is obtained
obtained from
from the
the SEL-735
SEL-735 m m in
in the
the field
field for
for control
control purposes.
purposes. The
The frequency
frequency signal
signal is
is passed
passed
through an outlier rejection filter to prevent overly large ramp rates and a low pass
through an outlier rejection filter to prevent overly large ramp rates and a low pass filter to preventfilter to prevent
potential oscillations.
potential oscillations.

Figure 4.
Figure Schematic of
4. Schematic of the
the frequency
frequency response
response control
control algorithm.
algorithm.

The primary
The primary component
component for for adjusting
adjusting thethe frequency
frequency response
response algorithm
algorithm is is the
the frequency-Watt
frequency-Watt
(f-W) curve. The curve can be configured with a deadband
(f-W) curve. The curve can be configured with a deadband of various widths and with of various widths and with various
various
proportional gains, i.e., the slope of the f-W curve in MW/Hz (Figure
proportional gains, i.e., the slope of the f-W curve in MW/Hz (Figure 4, inset). The result of the 4, inset). The result of the f-W
f-W
curve is
curve is summed
summed with with aa real
real power
powercomponent
componenttotomaintain
maintainaatarget targetstate-of-charge
state-of-charge(SOC).
(SOC).This
This sum
sum is
then limited by total available power, with a further limitation based on
is then limited by total available power, with a further limitation based on BMS and inverter data BMS and inverter data before
sending
before a real power
sending command
a real power to the inverter.
command to the inverter.
The frequency response algorithm
The frequency response algorithm was wasfirst
firsttested
tested within
within a simple
a simple gridgrid model
model developed
developed on
on the
the MATLAB/Simulink platform. A model of the BESS was incorporated
MATLAB/Simulink platform. A model of the BESS was incorporated into a simple model of the grid into a simple model of
the grid dynamics
dynamics [25], inawhich
[25], in which singlea generator
single generator
model model represented
represented the cumulative
the cumulative effecteffect
of allofthe
all
the generators on the grid, and which included a wind generation
generators on the grid, and which included a wind generation input (Figure 5). The unknown input (Figure 5). The unknown
parameters of
parameters of the
thegrid
gridmodel
modelwerewereinertia
inertiaconstant
constant (M), damping
(M), damping constant
constant(D),(D),
governor gains
governor (Kp (Kp
gains and
Ki), Ki),
and and andgenerator response
generator time.
response These
time. were
These estimated
were estimated based on historical
based on historicaltelemetry
telemetryprovided
provided by
HELCO. The consequence of a large number of unknowns is the
by HELCO. The consequence of a large number of unknowns is the possibility of parameter trades possibility of parameter trades
(i.e., aa particular
(i.e., particular setset of
of parameters
parameters causes
causes the
the model
model to to “fit”
“fit” the
the grid
grid frequency
frequency data,
data, but
but this
this may
may notnot
be the physically correct set of parameters). For this reason, a variety of reasonable
be the physically correct set of parameters). For this reason, a variety of reasonable parameter sets parameter sets were
used during
were testing.
used during This, in
testing. turn,inprovided
This, a variety
turn, provided of possible
a variety predicted
of possible behaviors.
predicted The full range
behaviors. The full of
resultsofindicated
range that the that
results indicated BESSthewould
BESSnot worsen
would the gridthe
not worsen frequency. Model runs
grid frequency. Model using
runsthe
usingsetsthe
of
telemetry-estimated parameters showed that the frequency response
sets of telemetry-estimated parameters showed that the frequency response algorithm would algorithm would improve both
the transient
improve bothresponse of theresponse
the transient grid to disturbances
of the grid and the steady-state
to disturbances and variability in gridvariability
the steady-state frequency.in
grid frequency.
During the site acceptance test, the grid frequency was observed for several 6–8 Hr periods with
the BESS off, and several 6–8 Hr periods with the BESS on. The results of the test showed that the
BESS did not degrade frequency variability, but there was also no statistically significant evidence
that the BESS reduced frequency variability with the proportional gain at 10 MW/Hz. Subsequent
testing by HNEI revealed that a statistically significant reduction in grid frequency variability could
be achieved by increasing the proportional gain (slope of the f-W curve) to 30 MW/Hz or more. Also,
because grid conditions and generator dispatch change over a shorter time frame, it was preferable
to make observations over shorter time periods rather than 6–8 Hr windows. A new protocol was
therefore developed for testing the BESS under the frequency response control algorithm,
Energies 2018, 11, 3367
as
7 of 17
described in the following section.

Figure 5.
Figure Schematic of
5. Schematic of the
the simple
simple model
model of
of the
the Hawaii
Hawaii island
island grid,
grid, which
which was
was used
used to
to develop
develop the
the
frequency response algorithm.
frequency response algorithm.

During the site acceptance test, the grid frequency was observed for several 6–8 Hr periods with
In order to test the effect of the BESS on grid frequency during field operations, HNEI designed,
the BESS off, and several 6–8 Hr periods with the BESS on. The results of the test showed that the BESS
and HELCO approved a test protocol to mitigate the effects of time varying background grid
did not degrade frequency variability, but there was also no statistically significant evidence that the
conditions. The protocol called for switching the BESS ON and OFF every 20 min allowing the
BESS reduced frequency variability with the proportional gain at 10 MW/Hz. Subsequent testing by
frequency variability of consecutive OFF/ON periods to be compared to assess the impact of the BESS.
HNEI revealed that a statistically significant reduction in grid frequency variability could be achieved
A time period of 20 min was chosen for each OFF and ON period in order to balance sufficient data
by increasing the proportional gain (slope of the f-W curve) to 30 MW/Hz or more. Also, because
collection against the changing background grid conditions.
grid conditions and generator dispatch change over a shorter time frame, it was preferable to make
The “switching experiments”, typically 200 min in duration, were initially performed only in the
observations over shorter time periods rather than 6–8 Hr windows. A new protocol was therefore
evening hours to isolate the frequency variability due to wind fluctuations. Examples of two evening
developed for testing the BESS under the frequency response control algorithm, as described in the
experiments are shown in Figure 6. One experiment was conducted during an evening with light
following section.
winds (Figure 6a) and another was conducted during an evening with medium winds (Figure 6b). In
In order to test the effect of the BESS on grid frequency during field operations, HNEI designed,
both, the red oscillating lines in Figure 6 (top) indicate when the battery was on showing real power
and HELCO approved a test protocol to mitigate the effects of time varying background grid conditions.
output of the BESS. The center plots show wind power output at the 10.6 MW Haw’i wind farm. The
The protocol called for switching the BESS ON and OFF every 20 min allowing the frequency variability
bottom plots display the grid frequency during the experiments. Comparing contiguous red and
of consecutive OFF/ON periods to be compared to assess the impact of the BESS. A time period of
black sections in the bottom plots, it is apparent that there was a reduction in grid frequency
20 min was chosen for each OFF and ON period in order to balance sufficient data collection against
variability when the BESS was ON (red) compared to when the BESS was OFF (black).
the changing background grid conditions.
The “switching experiments”, typically 200 min in duration, were initially performed only in the
evening hours to isolate the frequency variability due to wind fluctuations. Examples of two evening
experiments are shown in Figure 6. One experiment was conducted during an evening with light
winds (Figure 6a) and another was conducted during an evening with medium winds (Figure 6b).
In both, the red oscillating lines in Figure 6 (top) indicate when the battery was on showing real power
output of the BESS. The center plots show wind power output at the 10.6 MW Haw’i wind farm. The
bottom plots display the grid frequency during the experiments. Comparing contiguous red and black
sections in the bottom plots, it is apparent that there was a reduction in grid frequency variability
when the BESS was ON (red) compared to when the BESS was OFF (black).
Energies 2018,
Energies 11,11,
2018, 3367
x FOR PEER REVIEW 8 8ofof
17 17
Energies 2018, 11, x FOR PEER REVIEW 8 of 17

(a) (b)
(a) (b)
Figure 6. Top: BESS power output with a gain of 30MW/Hz; Middle: wind power generation; Bottom:
Figure6.6.Top:
Top:BESS
BESS power
power output
output with
with aa gain
gain of 30MW/Hz; Middle: wind power generation; Bottom:
Figure
and grid frequency on an evening with lightof 30MW/Hz;
winds (a) andMiddle: wind
an evening power
with generation;
medium winds Bottom:
(b). Black
and
and grid
grid frequency
frequency on
on anan evening
evening with
with light
light winds
winds (a)
(a) and
and an
an evening
evening with
with medium
medium winds
winds (b).
(b).Black
Black
lines indicate the BESS is OFF and red lines indicate BESS is ON. The BESS was switched between
lines
lines indicate
indicate the
the BESS
BESS is is OFF
OFF and
and red
red lines
lines indicate
indicate BESSBESS
is is
ON. ON.
TheThe BESS
BESS was was switched
switched between
between OFF
OFF and ON every 20 min.
OFFON
and andevery
ON every 20 min.
20 min.
Inorder
ordertotoquantify
Inorder
quantifythe thereduction
reductioninin grid frequency variability,
grid frequency
frequency variability,
the standard deviation of grid
In to quantify the reduction in grid variability, the
the standard
standarddeviation
deviationofofgridgrid
frequencywas
frequency wascalculated
calculatedfor foreach
each20-min
20-min period.
period. ToTo verify that this metric captured grid frequency
frequency was calculated for each 20-min period. To verify
verify that
that this
this metric
metriccaptured
capturedgrid
gridfrequency
frequency
variabilityinina abelievable
variability believableway, way,5 5sample
sampletime time series
series of of grid frequency were ordered by eye (left to
variability in a believable way, 5 sample time series of grid
grid frequency
frequency were
were ordered
orderedby byeye
eye(left
(lefttoto
right)
right)by by
byaafewa few researchers
fewresearchers (Figure
researchers (Figure
(Figure 7). 7). The
7). The ordering was then compared to the values of the metric.
right) The ordering
ordering was was then
then compared
comparedtotothethevalues
valuesofofthe
themetric.
metric.
The standard deviations (in mHz) were 7.1, 11.2, 14.2, 17.6, and 24.3 respectively from left to right in
The standard deviations (in mHz) were 7.1, 11.2, 14.2, 17.6, and 24.3 respectively from left to rightinin
The standard deviations (in mHz) were 7.1, 11.2, 14.2, 17.6, and 24.3 respectively from left to right
Figure 7, corresponding
Figure7,7,corresponding
correspondingto to
tothe the visual
thevisual ordering
visual ordering
ordering of of
of thethe frequency
the frequency variability.
frequency variability.
variability.
Figure

Figure
Figure 7.
Figure Five
7.7.
Five 20-min
Five frequency
20-min time
frequency series
series
time shown
shown
series inin
in
shown order
order of of
of
order increasing
increasing by-eye
by-eye
increasing variability.
variability.
by-eye variability.

Table
Table22shows
Table 2shows
shows the
the
thestandard
standard
standard deviations
deviations
deviations and percent
percent
and percent changes
changes
changes for
for the
the
for data
data
the data presented
presented
presented ininFigure
inFigure 6.6.6.
Figure
Each
Each OFF/ON
OFF/ON pair is referred to as an “interval”.
“interval”. The
The standard
standard deviations
deviations for
for the
the
Each OFF/ON pair is referred to as an “interval”. The standard deviations for the 5 intervals for the55 intervals
intervals for
for the
the
evening
evening
eveningwith
with
withlow
low
lowwind
wind
wind is shown
isisshown
shown in the
inin
thetheupper
upperpart
partofofthe table,
table,
the and
and
table, and the
the 55 intervals
the intervals
5 intervals forfor
for thethe
the evening
evening
evening
with
with medium
medium winds
winds is
is shown
shown in
in the
the bottom.
bottom. The
The mean
mean percent
percent change
change
with medium winds is shown in the bottom. The mean percent change for the light wind case for
for the
the light
light wind
wind case
case isis is
−−15.4%,
15.4%, with the negative indicating
indicating aa reduction
reduction of
of frequency
frequency variability
variability by
by 15.4%
15.4%
−15.4%, with the negative indicating a reduction of frequency variability by 15.4% when the BESS when
when the
the BESS
BESS isis is
Energies 2018, 11, x FOR PEER REVIEW 9 of 17

ON. The mean percent change for the medium wind case is −37.1%, indicating that with increasing
grid frequency variability, the BESS has a stronger proportional effect.
Although
Energies the above approach is a useful metric for assessing how the overall frequency
2018, 11, 3367 9 of 17
variability is reduced, it does not provide any information on how well the frequency variability is
reduced on short timescales, and therefore does not optimally measure the effect of using a fast-acting
ON. TheIf mean
battery. percent of
a timescales change
severalforminutes
the medium windwere
or longer case is −37.1%,
chosen, indicating
then the gridthat with increasing
frequency control
grid frequency variability, the BESS has a stronger proportional effect.
would be dominated by generator governors, and one is not measuring the effectiveness of the BESS.
Although
For this reason,the above
it was approach
decided thatisthe
a useful
20-minmetric for assessing
periods would behow the overall
divided frequency
into several variability
smaller non-
is reduced, it does not provide any information on how well the frequency
overlapping segments, and the standard deviation of each of the smaller segments taken. The mean variability is reduced on
short timescales, and therefore does not optimally measure the effect of
of those standard deviations is then determined and used to provide a single metric characterizing using a fast-acting battery.
If awhole
the timescales
20-min of several
period. minutes
The length or longer
of the were
smaller chosen, then is
segments thereferred
grid frequency
to as thecontrol would
timescale. Thisbe
dominated by generator governors, and one is not measuring the effectiveness
process is illustrated by example in Figure 8. The grid frequency data is separated into smaller 60 s of the BESS. For this
reason, it was
segments. The decided
standardthat the 20-min
deviations periods
of those 60would be divided
s segments into several
are shown smaller
just above the non-overlapping
plot. The mean
segments, and the standard deviation of each of the smaller segments
of those values was calculated to be 7.5 mHz and represents the variability for the entire 20-min taken. The meanwindow.
of those
standard deviations is then determined and used to provide a single metric characterizing the whole
20-min
Tableperiod. The length
2. The standard of theofsmaller
deviation segments
the 20-min intervalsis with
referred
BESS toOFFas versus
the timescale. This process
20-min periods with is
illustrated
BESS ON, byalong
examplewith in
theFigure
percent8.change
The grid frequency
between data
intervals, foristhe
separated
switchinginto smaller 60
experiments s segments.
shown in
The Figure
standard
6. A:deviations of those
Results for the evening 60with
s segments areB:shown
light winds; Resultsjust above
for an the with
evening plot.medium
The mean of those
winds.
values was calculated to be 7.5 mHz and represents the variability for the entire 20-min window.
A: Frequency Variability: Evening with Light Wind
Interval Standard Deviation of Standard Deviation of Percent Change
Table 2. The standard deviation of the 20-min intervals with BESS OFF versus 20-min periods with
Frequency w/ BESS OFF Frequency w/ BESS ON
BESS ON, along with the percent change between intervals, for the switching experiments shown in
[mHz]
Figure 6. A: Results for the evening [mHz]
with light winds; B: Results for an evening with medium winds.
1 13.4 8.3 −38.1
2 A: Frequency
10.3 Variability: Evening with6.0
Light Wind −41.7
3 Interval Standard 9.0
Deviation of Standard
7.7Deviation of Percent−14.4
Change
Frequency w/BESS OFF [mHz] Frequency w/BESS ON [mHz]
4 7.1 6.7 −5.6
1 13.4 8.3 −38.1
5 2 6.6
10.3 8.16.0 22.7
−41.7
3 B: Frequency9.0
Variability: Evening with Medium
7.7 Wind −14.4
4 7.1 6.7 −5.6
Interval
5
Standard Deviation
6.6
of Standard Deviation
8.1
of Percent Change
22.7
Frequency w/ BESS OFF Frequency w/ BESS ON
B: Frequency Variability: Evening with Medium Wind
[mHz] [mHz]
Standard Deviation of Standard Deviation of
1 Interval 15.9 OFF [mHz]
Frequency w/BESS 8.6
Frequency w/BESS ON [mHz]
Percent−45.9
Change

2 1 13.5
15.9 7.98.6 −41.5
−45.9
3 2 13.5
14.3 8.87.9 −41.5
−38.5
3 14.3 8.8 −38.5
4 4 12.6
12.6 7.27.2 −42.9
−42.9
5 5 11.4
11.4 9.59.5 −16.7
−16.7

Figure 8. A frequency time-series is divided into several smaller segments, in this case 60-seconds in
length. The standard deviation of each smaller time series is shown as just above the plot. The mean of
those values is taken to be the “frequency variability” of the 20-min period on a 60-seconds timescale.
Energies 2018, 11, x FOR PEER REVIEW 10 of 17

Figure 8. A frequency time-series is divided into several smaller segments, in this case 60-seconds in
length. The standard deviation of each smaller time series is shown as just above the plot. The mean
of those values is taken to be the “frequency variability” of the 20-min period on a 60-seconds
Energies 2018, 11, 3367 10 of 17
timescale.

Timescales
Timescales thatthatare
aretoo
tooshort
shortpresent
presentproblems
problems as as well.
well.Figure
Figure9,9,shows
showsthethemean
meanof ofthe
thestandard
standard
deviations
deviations over a number of timescales and for the same samples as shown in Figure 7,the
over a number of timescales and for the same samples as shown in Figure 7, with same
with the
colors (e.g., Sample 1 is blue in both figures). At timescales from seconds to
same colors (e.g., Sample 1 is blue in both figures). At timescales from seconds to tens of seconds,tens of seconds, the
measured standard
the measured deviation
standard depends
deviation on theon
depends chosen time scale,
the chosen timeso one so
scale, does
onenot have
does a stable
not have ametric.
stable
Moreover, the shortest
metric. Moreover, timescales
the shortest do notdoshow
timescales goodgood
not show separation between
separation betweenthethe
high
high andandlow
lowgrid
grid
variability periods. At timescales of approximately one minute and longer,
variability periods. At timescales of approximately one minute and longer, the measured standardthe measured standard
deviation
deviationvalues
valueslevel
leveloff
offtowards
towardsa near-constant
a near-constant value; in in
value; other words,
other words,thethe
measured
measured values are no
values are
longer timescale dependent. Moreover, the separation between more and
no longer timescale dependent. Moreover, the separation between more and less variable periods less variable periods
increases
increasesupupto tothe
theone-minute
one-minutetimescale,
timescale,and andthen
thenlevels
levelsoff
offasas well.
well. The
Theone-minute
one-minutetimescale
timescalewas
was
therefore
therefore chosen to balance the need to capture the quick BESS response with stable metricthat
chosen to balance the need to capture the quick BESS response with stable metric thatcan
can
distinguish high a low variability
distinguish high a low variability periods.periods.

Figure
Figure 9.9. Frequency
Frequencyvariability
variabilitymetrics
metricsfor
fordata
datashown
shownin
in Figure
Figure 7,
7, using
using different
different timescales.
timescales. The
The
colors
colorscorrespond
correspondto tothe
thesamples
samplesshown
shownininFigure
Figure 7.
7.

With
Withthese
theseresults,
results,the
thechosen
chosen metric to to
metric characterize
characterize gridgrid
frequency
frequency (whether the BESS
(whether was ON
the BESS was
or OFF) was the mean of consecutive 60-seconds standard deviations. With
ON or OFF) was the mean of consecutive 60-seconds standard deviations. With this metric, it is this metric, it is possible
to study the
possible BESS the
to study performance under various
BESS performance under grid conditions
various and control
grid conditions and algorithm settings.settings.
control algorithm Initial
results
Initial of switching
results experiments
of switching are shown
experiments arein Figurein10,
shown which
Figure 10,shows
which40-min
showstime40-minintervals of BESS
time intervals
real power
of BESS output
real power (top),
outputHRD wind
(top), power
HRD windoutput
power (2nd from(2nd
output top),from
grid top),
frequency (2nd from(2nd
grid frequency bottom),
from
and the running
bottom), standard
and the running deviation
standard (60-seconds
deviation timescale)
(60-seconds of grid
timescale) frequency
of grid frequency (bottom),
(bottom),for fortwo
two
switching experiment performed
switching experiment performedunder under similar
similar gridgrid conditions,
conditions, butdifferent
but with with different
frequency frequency
response
response control algorithm
control algorithm settings. Specifically,
settings. Specifically, the left side the leftfigure
of the side shows
of the anfigure shows an
experiment experiment
conducted with
conducted
the full 1 MWwith BESS
the full 1 MWoutput.
power BESS power output. Theon
The experiment experiment
the right on theofright
side side ofwas
the figure the figure was
conducted
conducted with the
with the power power
output output
limited limited
to 300 kW.toThe300winds
kW. The were winds werehigh
similarly similarly high and
and variable invariable in
both cases.
both cases. The bottom-left
The bottom-left plot clearlyplot clearly
shows thatshows
the BESSthatOFF
the BESS
periods OFF periods
(black) have(black)
largerhave largerdeviations
standard standard
deviations
that the BESSthatONtheperiods
BESS ON (red).periods
This is (red). This is in
less apparent less
theapparent
plot on thein bottom-right
the plot on the bottom-right
because the BESS
because
is beingthe BESS istobeing
restricted 300 kW.restricted to 300 kW.
Energies 2018, 11, 3367 11 of 17
Energies 2018, 11, x FOR PEER REVIEW 11 of 17

(a) (b)
Figure
Figure 10.10.
AA comparisonofoftwo
comparison twodifferent
differentdays
days when
when total
total wind
wind power
poweroutput
outputatatHaw’i
Haw’iRenewable
Renewable
Development (HRD) and variability were similar. The gain setting used for the f-W curve of the BESS
Development (HRD) and variability were similar. The gain setting used for the f-W curve of the BESS
control algorithm was 30MW/Hz in both cases. The only difference is the limiting of the BESS output
control algorithm was 30MW/Hz in both cases. The only difference is the limiting of the BESS output
power, set to 1 MW for the left plots and 300 kW for the right plots. (Top) Time series of BESS real
power, set to 1 MW for the left plots and 300 kW for the right plots. (Top) Time series of BESS real
power output. (2nd from top) Time series of wind power at the HRD wind farm. (2nd from bottom)
power output. (2nd from top) Time series of wind power at the HRD wind farm. (2nd from bottom)
Time series of grid frequency. (bottom) The 60-seconds standard deviation of grid frequency.
Time series of grid frequency. (bottom) The 60-seconds standard deviation of grid frequency.
Table3 3lists
Table liststhe
the60-seconds
60-seconds frequency
frequency variability
variability metrics
metrics associated
associatedwith
withthe
theswitching
switching
experiments shown in Figure 10. The average percent variability reduction for
experiments shown in Figure 10. The average percent variability reduction for the 1 MW the 1 MW case (left)
case is is
(left)
58.4 ± 15% with a 95% confidence interval, and the average percent reduction for the 300-kW case
58.4 ± 15% with a 95% confidence interval, and the average percent reduction for the 300-kW case
(right) is 20.9 ± 11% with a 95% confidence interval. The indication is that limiting the BESS output to
(right) is 20.9 ± 11% with a 95% confidence interval. The indication is that limiting the BESS output
30% of rated power yields about a linear 30% reduction in frequency variability for similar wind
to 30% of rated power yields about a linear 30% reduction in frequency variability for similar wind
conditions. The linear relationship was later confirmed through experiments designed to examine of
conditions. The linear relationship was later confirmed through experiments designed to examine
the effect of various frequency response control algorithm parameter settings on BESS frequency
of the effect of various frequency response control algorithm parameter settings on BESS frequency
regulation and usage [20].
regulation and usage [20].
Table 3. The 60-seconds frequency variability metric for the 20-min periods with BESS OFF against
Table 3. The 60-seconds frequency variability metric for the 20-min periods with BESS OFF against
adjacent 20-min periods with BESS ON show in Figure 10. A: Results for the evening with high winds
adjacent 20-min periods with BESS ON show in Figure 10. A: Results for the evening with high winds
and the BESS allowed to use full rated power of 1 MW; B: Results for the evening with similar high
and the BESS
winds allowed
and the to use full
BESS allowed rated
to use 300power
kW of of 1 MW; B: Results for the evening with similar high
power.
winds and the BESS allowed to use 300 kW of power.
A: 60-Second Frequency Variability: 1000 kW Limit
Interval A: 60-Second
Frequency Frequency
Metric w/ Variability:
Frequency1000 kW Limit
Metric w/ Percent Change
Frequency
BESS OFF Metric
[mHz] Frequency
BESS ON Metric
[mHz]
Interval Percent Change
w/BESS OFF [mHz] w/BESS ON [mHz]
1 12.9 7.4 −74.3
1 12.9 7.4 −74.3
2 10.7 7.2 −48.6
2 10.7 7.2 −48.6
3 3 11.2
11.2 6.7
6.7 −67.2 −67.2
4 4 10.9
10.9 7.0
7.0 −55.7 −55.7
5 10.8 7.4 −45.9
5 10.8 7.4 −45.9
B: 60-Second Frequency Variability: 300
B: 60-Second Frequency Variability: 300 kW LimitkW Limit

IntervalInterval Frequency
Frequency Metric
Metric w/ Frequency Metric
Frequency Metric w/ Percent Percent
Change Change
w/BESS OFF [mHz] w/BESS ON [mHz]
BESS OFF [mHz] BESS ON [mHz]
1 9.2 7.0 −23.9
1 2
9.2
11.5 8.5
7.0 −26.1
−23.9
2 3 11.5
12.7 8.5
8.9 −29.9 −26.1
3 4 11.2
12.7 10.5
8.9 −6.3 −29.9
5 12.1 9.9 −18.2
Energies 2018, 11, x FOR PEER REVIEW 12 of 17

Energies 2018, 11, 3367 12 of 17

4 11.2 10.5 −6.3
5 12.1 9.9 −18.2
3.2. Wind Smoothing: Algorithm Development and Testing
3.2. Wind Smoothing: Algorithm Development and Testing
In contrast with the frequency response algorithm, where switching experiments were imperative,
only In contrastpassive
standard with the frequencywere
observations response
required algorithm, where the
to characterize switching experiments
performance were
of the wind
imperative, only standard passive observations were required to characterize the
smoothing algorithm. This section is not intended to show novelty but was included in this report performance of the
wind smoothing algorithm. This section is not intended to show novelty but was included in this
for completeness.
reportTheforHRD
completeness.
wind farm that came online in 2006, consists of 16 Vestas American Wind Technology
The HRD wind
660 kW wind turbines, farm that
for camerated
a total online in 2006,ofconsists
capacity 10.56 MWof 16[26].
Vestas American
This wind plantWind Technology
and purchase
660 kW wind turbines, for a total rated capacity of 10.56 MW [26]. This wind
power agreement pre-dated the suite of modern wind plant capabilities available today, which plant and purchase
include
power agreement pre-dated the suite of modern wind plant capabilities available
frequency response; primary frequency response would leverage the capabilities of the wind turbine today, which
include
controls frequency
to respondresponse; primary
naturally to frequency
negative response impacts
system frequency would leverage the capabilities
from its volatility. of the wind
The requirements
turbine controls to respond naturally to negative system frequency impacts
for the wind power smoothing algorithm were developed based on the Power Purchasing Agreement from its volatility. The
requirements for the wind power smoothing algorithm were developed based
(PPA) between the utility (HELCO) and the wind farm owner, which includes three limitations on the on the Power
Purchasing
fluctuationsAgreement
of the wind(PPA) between
farm power the utility
output. (HELCO)
Specifically, and
the PPAthestipulates
wind farm owner, which includes
that:
three limitations on the fluctuations of the wind farm power output. Specifically, the PPA stipulates that:
• the wind farm’s power output may not change more than 1 MW over any two seconds interval,
• the wind farm’s power output may not change more than 1 MW over any two seconds interval,
• the wind farm’s power output may not change more than 2 MW over any one-minute interval,
• the wind farm’s power output may not change more than 2 MW over any one-minute interval,
•
• the
the one-minute
one-minute average
average of power output
of power output changes
changes over
over any
any 22 ss intervals
intervals may
may not
not exceed
exceed 300
300 kW.
kW.

A simplified
A simplified block
blockdiagram
diagramofofthethe wind
wind power
power smoothing
smoothing algorithm
algorithm is shown
is shown in Figure
in Figure 11. The11.
The algorithm was designed to minimize PPA agreement violations, as well
algorithm was designed to minimize PPA agreement violations, as well as minimize the standard as minimize the standard
deviation of
deviation of power
power output
output over
over the
the 2-seconds
2-seconds and
and one-minute
one-minute intervals.
intervals. InIn the
the most
most basic
basic terms,
terms, the
the
algorithm calculates a “target” power output from the wind farm based on the
algorithm calculates a “target” power output from the wind farm based on the filtered values of the filtered values of the
“raw”
“raw”windwindfarmfarmoutput
output overover
an adjustable periodperiod
an adjustable of time.ofAtime.
standard proportional-integral-derivative
A standard proportional-integral-
(PID) controller then calculates the desired BESS power command
derivative (PID) controller then calculates the desired BESS power command to offsetto offset the difference between
the difference
the actual wind farm output and that target value. All of the parameters were
between the actual wind farm output and that target value. All of the parameters were first tuned first tuned to avoidto
oscillations
avoid and then
oscillations optimized
and then using
optimized previous
using SCADA
previous SCADA data
dataof of
HRD
HRDwind windfarmfarm power output.
power output.
Additionally,
Additionally, thethe algorithm
algorithm was was designed
designed to to return
return the
the BESS
BESS toto the
the nominal
nominal 50% 50% state
state of
of charge
charge by by
calculating the battery “bias”, which is defined as the amount of power needed
calculating the battery “bias”, which is defined as the amount of power needed to return the BESS to to return the BESS
to 50%
50% SOCSOC from
from itsits currentSOC
current SOCover
over1515min.
min.Because
Becausethe
thebattery
batterybias
bias is
is continuously
continuously updated,
updated, thethe
associated bias power commands exponentially approach zero as the battery
associated bias power commands exponentially approach zero as the battery nears 50% SOC, with nears 50% SOC, with the
result
the being
result thatthat
being the SOC
the SOCmanagement
management could taketake
could the battery fromfrom
the battery a fulla charge or zero
full charge charge
or zero to 50%
charge to
SOC in approximately one hour. The final algorithm was validated against the
50% SOC in approximately one hour. The final algorithm was validated against the subset of SCADA subset of SCADA data
that was
data thatnot
wasusednot for
used algorithm tuning.tuning.
for algorithm The algorithm passedpassed
The algorithm laboratory and siteand
laboratory acceptance tests to
site acceptance
the satisfaction
tests of HNEIoftechnical
to the satisfaction reviewers.
HNEI technical reviewers.

Figure 11.
Figure Schematic of
11. Schematic of the
the wind
wind smoothing
smoothing control
control algorithm.
algorithm.

Figure 12
Figure 12 shows
showsa atime
timeseries
series
of of power
power output
output from from the HRD
the HRD windwind
farm farm
(black)(black)
and theand the
BESS-
BESS-smoothed
smoothed time (red),
time series seriesfor
(red),
the for the morning
morning of Aprilof4th,April
2013.4th,
The2013.
bottom The bottom
three plotsthree plots are
are expanded
expanded sections of the time series to show detail. Qualitatively, it is apparent from comparison
sections of the time series to show detail. Qualitatively, it is apparent from comparison of the two of
the two time series (black and red) that the algorithm is able to smooth output from
time series (black and red) that the algorithm is able to smooth output from the wind farm. the wind farm.
 Energies
Energies
Energies 2018,
2018,
2018, 11,
11,
11, x FOR
x3367
FOR PEER
PEER REVIEW
REVIEW 13of
1313 ofof17
1717

Figure1313shows
Figure showshistograms
histogramsofofmetricsmetricsrelated
relatedtotothe thethree
threePPAPPAlimitations
limitationscalculated
calculatedfrom fromthe the
BESS
BESS dataset
Figure
dataset spanning
13spanning 8 8a.m.
shows histograms a.m.toto6of
6p.m.,
p.m.,
metricsApril
April 5th2013.
related
5th 2013.
to the The
The three three
three PPA metrics
limitations
metrics are,from
are, from lefttotoright:
calculated
left right:the
from the
maximum change in power output over two seconds, the maximum change in power output over 6060
maximum
BESS dataset change
spanning in power
8 a.m. output
to 6 over
p.m., two
April seconds,
5th 2013. the
The maximum
three change
metrics are,in power
from left output
to over
right: the
s,and
andthe
maximum
s, theone-minute
one-minute
change in poweraverages
averages ofofthe
output themaximum
over maximum
two seconds, changes
changes the overovertwo
maximum twoseconds.
seconds. inAgain,
change Again, power thethe windfarm
output
wind farm
over
60power
powers, and output isisrepresented
the one-minute
output represented
averagesbybyof the
the black
theblack
maximum lineand
line andthe
changes thesmoothed
smoothed
over two seconds.powerby
power bythe
Again, thethe red
red windline.farm
line. The
The
histograms
power output show
is that
represented the wind
by the smoothing
black line andalgorithm
the is
smoothed very
histograms show that the wind smoothing algorithm is very effective at reducing power output effective
power by the at reducing
red line. The power output
histograms
fluctuations
show that the
fluctuations atat
wind
thetheone-minute
one-minute timescalebut
smoothingtimescale
algorithm but
is isissignificantly
very significantly
effective atless lesseffective
effective
reducing poweratatthe
the twoseconds
output
two secondstimescale.
fluctuations timescale.
at the
At the fast two seconds timescale, the algorithm reduces range of
At the fast two seconds timescale, the algorithm reduces the range of the maximum change (left), two
one-minute timescale but is significantly less effective at the two secondsthe maximum
timescale. change
At the (left),
fast but
but
themean
seconds
the mean valueremains
timescale,
value remains virtually
the algorithm
virtually unchanged
reduces
unchanged (right).
the range
(right). However,
of However,
the maximum atatthe
the one-minute
change
one-minute(left), buttimescale
the mean
timescale (center),
value
(center),
thealgorithm
remains
the algorithm
virtually
hashas a astrong
strongimpact,
unchanged impact, withthe
(right).with theaverage
However, average maximumchange
at the maximum
one-minute changeininpower
timescale poweroutput
(center), output reduced
the algorithm
reduced byby
has
a more
strong than
more than 60%. 60%.
impact, with the average maximum change in power output reduced by more than 60%.

Figure12.
Figure
Figure 12.Time
Time
Time series
series
series ofof
of power
power
power output
output
output from
from
from theHRD
thethe
HRD HRD wind
wind
wind farm
farm
farm (black)
(black)
(black) andand and
the thesmoothed
the smoothed
smoothed power
power
power (red)
(red)
for the
(red) for the
formorningmorning
the morning of 4/4/2013.
of 4/4/2013.
of 4/4/2013. The
TheThe bottom
bottom
bottomthreethree
plots
three plots
show
plots show the
thethe
show sections
sections indicated
indicated
sections indicatedbyby by the
thethe
blue blue boxes
boxes
blue boxesin
ingreater
greater
greater
in detail.
detail.
detail.

Figure13.
Figure
Figure 13.Histograms
13. Histogramsof
Histograms ofofthree
threemetrics
three metricsrelated
metrics relatedto
related totothe
thepower
the powerpurchasing
power purchasingagreement
purchasing agreementbetween
agreement betweenthe
between theHRD
the HRD
HRD
windfarm
wind
wind farmand
farm andthe
and theutility,
the utility,HELCO.
utility, HELCO.The
HELCO. Themetrics
The metricsare
metrics arethe
are themaximum
the maximumchange
maximum changein
change ininpower
poweroutput
power outputover
output over22 2ss s
over
(left),
(left), the maximum change in power output over 60 s (middle), and
(left), the maximum change in power output over 60 s (middle), and the one-minute average of
the maximum change in power output over 60 s (middle), and the
the one-minute
one-minute average
average ofofthe
the
the
maximum
maximum power
power output
output over
over 2 2
s s (right).
(right). The
The black
black (red)
(red) lines
lines indicate
indicate metrics
metrics
maximum power output over 2 s (right). The black (red) lines indicate metrics calculated from data calculated
calculated from
from data
data
whenthe
when
when theBESS
the BESSwas
BESS wasinactive
was inactive(active).
inactive (active).The
(active). Thedata
The datawas
data wascollected
was collectedfrom
collected from88 8a.m.
from a.m.to
a.m. toto66 6p.m.
p.m.on
p.m. onon4/5/2013.
4/5/2013.
4/5/2013.

3.3.Early
3.3. EarlyTesting
Testingand
andLessons
LessonsLearned
Learned
Shortlyafter
Shortly aftercommissioning,
commissioning,the thefocus
focusofofthe
theresearch
researchwaswasthe
theoptimization
optimizationofofthe thefrequency
frequency
responsealgorithm.
response algorithm.Particular
Particularattention
attentionwas
waspaid
paidtotothe
the“gain”
“gain”setting
settingofofthe
thef-W
f-Wcurve.
curve.While
Whilehigher
higher
Energies 2018, 11, 3367 14 of 17

3.3. Early Testing and Lessons Learned

Shortly after commissioning, the focus of the research was the optimization of the frequency
Energies 2018, 11, x FOR PEER REVIEW 14 of 17
response algorithm. Particular attention was paid to the “gain” setting of the f-W curve. While higher
proportional gains led to greater grid benefit, it did so with a substantial burden on the BESS resulting
proportional gains led to greater grid benefit, it did so with a substantial burden on the BESS resulting
in more energy throughput (cycling). During early testing, gains of up to 40 MW/Hz were attempted
in more energy throughput (cycling). During early testing, gains of up to 40 MW/Hz were attempted
but operation for extended times at these higher gains were found to increase the temperature of the
but operation for extended times at these higher gains were found to increase the temperature of the
modules (each module is 1/2 of a line replaceable unit), as shown in Figure 14. As the temperature
modules (each module is 1/2 of a line replaceable unit), as shown in Figure 14. As the temperature
excursions would cause excessive wear on the battery modules, a setting of 40 MW/Hz with no
excursions would cause excessive wear on the battery modules, a setting of 40 MW/Hz with no
deadband was determined to be beyond the desirable operational limits of the BESS.
deadband was determined to be beyond the desirable operational limits of the BESS.
Another item of interest was the distribution of temperatures in the PM. It was found that
Another item of interest was the distribution of temperatures in the PM. It was found that the
the initial configuration of the HVAC air vents, set as they were upon shipment, were not optimal.
initial configuration of the HVAC air vents, set as they were upon shipment, were not optimal. In
In particular, the air vents shown in Figure 15 at the bases of the two HVAC units were wide open.
particular, the air vents shown in Figure 15 at the bases of the two HVAC units were wide open. This
This resulted in less air flow to the battery modules, which resulted in higher battery temperatures
resulted in less air flow to the battery modules, which resulted in higher battery temperatures overall.
overall. Further, this resulted in increased temperature gradient across the LRUs, as can be seen in
Further, this resulted in increased temperature gradient across the LRUs, as can be seen in Figure 15
Figure 15 (left). HNEI personnel made adjustments on-site, resulting in an improvement to both total
(left). HNEI personnel made adjustments on-site, resulting in an improvement to both total
temperature and temperature gradient (right). Note that the two sets of temperature measurements
temperature and temperature gradient (right). Note that the two sets of temperature measurements
shown in Figure 15 were collected on dates when weather and grid conditions, and the corresponding
shown in Figure 15 were collected on dates when weather and grid conditions, and the corresponding
BESS usage, were similar. There are a number of other studies currently underway that utilize data
BESS usage, were similar. There are a number of other studies currently underway that utilize data
from the subject BESSs. A recently published paper used BESS usage data over a three year period to
from the subject BESSs. A recently published paper used BESS usage data over a three year period to
develop a representative duty cycle [22], which was used to test cell degradation and calendar aging
develop a representative duty cycle [22], which was used to test cell degradation and calendar aging
in the laboratory [21].
in the laboratory [21].

Figure14.
Figure Timeseries
14.Time series of
of the temperature
temperature of ofbattery
batterymodules
modules(two
(twogroups
groupsofof
seven
sevencells) over
cells) a three-day
over a three-
period
day when
period the the
when gaingain
setting of the
setting frequency
of the control
frequency algorithm
control algorithm was increased
was increasedfrom
from10 10
MW/Hz
MW/Hz to
◦ ◦
to4040
MW/Hz.
MW/Hz.TheTheoperating
operatingspecification
specificationfor
forthese
thesebattery
batterycells between−−40
cellsisisbetween 40 °CC and
and +55 C.
+55 °C.
Energies 2018, 11, 3367 15 of 17
Energies 2018, 11, x FOR PEER REVIEW 15 of 17

Figure15.
15. Temperature
Temperature of
of the
the battery
battery modules
modules (two
(two groups
groups of
of seven
seven cells,
cells, ½
1 of a LRU) for different
Figure 2 of a LRU) for different
configuration of the HVAC1 vents. Units are degrees Celsius.
configuration of the HVAC1 vents. Units are degrees Celsius.

4.
4. Summary
Summary and
and Future
Future Work
Work
In
InDecember
Decemberofof2012, 2012,a 1a MW,
1 MW, 250250 kW-HrkW-Hrfast-acting
fast-actingBESSBESS
unit was
unitinstalled on the on
was installed Hawaii Island
the Hawaii
electric grid at the
Island electric gridtransmission level. Thislevel.
at the transmission BESSThis
was BESS the first
wasof the
three grid-scale
first of threeBESS units installed
grid-scale BESS unitsby
the Hawaiiby
installed Natural EnergyNatural
the Hawaii InstituteEnergy
as part of an integrated
Institute as partresearch, testing, and
of an integrated evaluation
research, program.
testing, and
Two control program.
evaluation algorithms were
Two developed,
control one designed
algorithms to smoothone
were developed, power outputtofrom
designed a co-located
smooth power outputwind
farm
fromand a second to
a co-located regulate
wind farmisland-wide
and a second gridtofrequency.
regulate Both control algorithms
island-wide proved
grid frequency. successful
Both control
at implementing
algorithms proved their control objectives.
successful at implementingWhen controlled
their control by the wind smoothing
objectives. algorithm,
When controlled by the
the BESS
wind
was able to algorithm,
smoothing reduce thethe maximum
BESS was change
able to inreduce
power the output from the
maximum windinfarm
change powerover a one-minute
output from the
periods
wind farm by more
over athan 60%. When
one-minute controlled
periods by more bythan
the frequency
60%. When response
controlled algorithm, the BESS response
by the frequency reduced
the average the
algorithm, standard
BESS deviation
reduced the of grid frequency
average standard overdeviation
one-minute of periods by between
grid frequency over20% and 60%
one-minute
compared
periods bytobetween
periods20% when and the60%
BESS was OFF,
compared todepending
periods when on thethe settings
BESS was forOFF,
the control
depending algorithm.
on the
The results
settings fordemonstrate that a relatively
the control algorithm. small BESS
The results unit, namely
demonstrate that aa1relatively
MW BESSsmall installed
BESSonunit,a grid with
namely
aa peak
1 MWload BESS of installed
around 180 on aMW, gridwas
withable to provide
a peak load of measurable
around 180 MW, grid benefit
was able and
to the utility
provide company
measurable
elected to develop
grid benefit and the a budget to maintain
utility company and operate
elected to develop the asystem.
budget to maintain and operate the system.
Additionally,
Additionally,the theresults
resultsdemonstrate
demonstratethat thatthetheimpact
impactof ofBESS
BESSunits
unitson onthe
thegrid,
grid,and
andindeed
indeed thethe
metric
metric used
used to to quantify
quantify the the impact,
impact, can can bebe highly
highly timescale
timescale dependent.
dependent. For the the case
case of of the
the wind
wind
smoothing
smoothing algorithm,
algorithm, while while thethe BESS
BESS was was able
able to
to significantly
significantly reducereduce wind
wind power
power output
output variability
variability
on
on the
the one-minute
one-minute timescale,
timescale, there
there was
wasalmost
almostno noeffect
effectat atthe
thetwo-seconds
two-secondstimescale.
timescale. TheThe measured
measured
effect
effectofofthe
theBESS
BESS on on grid
grid frequency
frequency whilewhile running
running the the frequency
frequency response
responsealgorithm
algorithmdepends
dependson on the
the
timescale
timescale thatthat isis examined.
examined. While it it is
is desirable
desirable to to measure
measure the the BESS
BESS impact
impact onon grid
grid frequency
frequency on on
short
shorttimescales
timescalesbecause
becauseof ofthethe quick
quick response
responsetime timeof ofthe
the BESS,
BESS, itit was
was shown
shown that
that timescales
timescales shorter
shorter
than
than aa minute
minute can can result
result inin measurements
measurementsof of grid
grid frequency
frequencythat thatarearetimescale
timescaledependent
dependentand anddo donot
not
separate
separate highhigh and
and lowlow variability
variability periods
periods well.well. ItIt may
may be be possible
possible to to take
take advantage
advantage of of the
the timescale
timescale
dependence
dependence to to optimize
optimize BESS BESS control
control algorithms
algorithms for for their
theirparticular
particularintended
intendedtasks.
tasks. TheThe natural
natural
timescale
timescalefor forsuch
suchBESS BESSoptimization
optimization would be at the
would be relatively short timescales
at the relatively over whichover
short timescales traditional
which
generation
traditional units are incapable
generation units areofincapable
responding. of responding.
Future work plans include investigating the impact of the Hawaii island BESS on large frequency
events, and the ability of a 2 MW, 397 kW-Hr BESS to help stabilize the small isolated grid on the
island of Molokai.
Energies 2018, 11, 3367 16 of 17

Future work plans include investigating the impact of the Hawaii island BESS on large frequency
events, and the ability of a 2 MW, 397 kW-Hr BESS to help stabilize the small isolated grid on the island
of Molokai.

Author Contributions: Conceptualization, R.R. and K.M.; Methodology, R.R., K.M., M.T., K.S.; Software, K.M.,
M.T., K.S.; Validation, K.M., and K.S.; Formal Analysis, K.M. and M.T.; Investigation, R.R.; Resources, R.R.; Data
Curation, M.T. and K.S.; Writing-Original Draft Preparation, K.S.; Writing-Review & Editing, R.R., M.T., and K.S.;
Visualization, M.T.; Supervision, R.R.; Project Administration, R.R.; Funding Acquisition, R.R.
Funding: This work was funded by ONR under the Hawaii Energy and Environmental Technologies (HEET) 2010
Initiative, award No N00014-11-1-0391 and the Asia Pacific Research Initiative for Sustainable Energy Systems
2012, award No. N00014-13-1-0463.
Acknowledgments: The authors are grateful to the Hawaii Electric Light Company for their support and
partnership on this project and their ongoing support to the operations of the Hawaii Sustainable Energy Research
Facility (HiSERF), Haw’i Renewable Development for hosting the BESS, and Jerry Haverstick and Len Sekowski
(Altairnano) for their help through the course of this study.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Eftekharnejad, S.; Vittal, V.; Heydt, G.T.; Keel, B.; Loehr, J. Impact of increased penetration of photovoltaic
generation on power systems. IEEE Trans. Power Syst. 2013, 28, 893–901. [CrossRef]
2. Ulbig, A.; Borsche, T.S.; Andersson, G. Impact of low rotational inertia on power system stability and
operation. arXiv, 2013, arXiv:1312.6435.
3. Georgilakis, P.S. Technical challenges associated with the integration of wind power into power systems.
Renew. Sustain. Energy Rev. 2008, 12, 852–863. [CrossRef]
4. Hamsic, N.; Schmelter, A.; Mohd, A.; Ortjohann, E.; Schultze, E.; Tuckey, A.; Zimmermann, J. Increasing
renewable energy penetration in isolated grids using a flywheel energy storage system. In Proceedings of the
International Conference on Power Engineering, Energy and Electrical Drives, POWERENG 2007, Setubal,
Portugal, 12–14 April 2007; pp. 195–200.
5. Hawaii Clean Energy Initiative. Available online: http://www.hawaiicleanenergyinitiative.org/ (accessed
on 27 November 2018).
6. Rampazzoa, M.; Luvisottoa, M.; Tomasonea, N.; Fastellib, I.; Schiavettib, M. Modelling and simulation of a
Li-ion energy storage system: Case study from the island of Ventotene in the Tyrrhenian Sea. J. Energy Storage
2018, 15, 57–68. [CrossRef]
7. El-Bidairia, K.; Nguyena, H.; Jayasinghea, S.; Mahmoudb, T.; Penesisa, I. A hybrid energy management
and battery size optimization for standalone microgrids: A case study for Flinders Island, Australia.
Energy Convers. Manag. 2018, 175, 192–212. [CrossRef]
8. Rodrigues, E.; Godina, R.; Catalão, J. Modelling Electrochemical Energy Storage Devices in Insular Power
Network Applications Supported on Real Data. Appl. Energy 2017, 188, 315–329. [CrossRef]
9. Cabrera, P.; Lund, H.; Carta, J. Smart renewable energy penetration strategies on islands: The case of Gran
Canaria. Energy 2018, 162, 421–443. [CrossRef]
10. Notton, G. Importance of Islands in Renewable Energy Production and Storage: The Situation of the French
Islands. Renew. Sustain. Energy Rev. 2015, 47, 260–269. [CrossRef]
11. Dunn, B.; Kamath, H.; Tarascon, J.-M. Electrical energy storage for the grid: A battery of choices. Science
2011, 334, 928–935. [CrossRef] [PubMed]
12. Koohi-Kamali, S.; Tyagi, V.; Rahim, N.; Panwar, N.; Mokhlis, H. Emergence of energy storage technologies as
the solution for reliable operation of smart power systems: A review. Renew. Sustain. Energy Rev. 2013, 25,
135–165. [CrossRef]
13. Beaudin, M.; Zareipour, H.; Schellenberglabe, A.; Rosehart, W. Energy storage for mitigating the variability
of renewable electricity sources: An updated review. Energy Sustain. Dev. 2010, 14, 302–314. [CrossRef]
14. Castillo, A.; Gayme, D.F. Grid-scale energy storage applications in renewable energy integration: A survey.
Energy Convers. Manag. 2014, 87, 885–894. [CrossRef]
Energies 2018, 11, 3367 17 of 17

15. Hidalgo-León, R.; Siguenza, D.; Sanchez, C.; León, J.; Jácome-Ruiz, P.; Wu, J.; Ortiz, D. A survey of battery
energy storage system (BESS), applications and environmental impacts in power systems. In Proceedings of
the Ecuador Technical Chapters Meeting (ETCM), Salinas, Ecuador, 16–20 October 2017; pp. 1–6.
16. Hernández, J.; Gyuk, I.; Christensen, C. DOE global energy storage database—A platform for large scale
data analytics and system performance metrics. In Proceedings of the 2016 IEEE International Conference
on Power System Technology (POWERCON), Wollongong, NSW, Australia, 28 September–1 October 2016;
pp. 1–6.
17. Consiglio, L.; Di Lembo, G.; Noce, C.; Eckert, P.; Rasic, A.; Schuette, A. Performances of the first electric
storage system of ENEL Distribuzione. In Proceedings of the AEIT Annual Conference 2013, Mondello, Italy,
3–5 October 2013.
18. Koller, M.; Borsche, T.; Ulbig, A.; Andersson, G. Review of grid applications with the Zurich 1 MW battery
energy storage system. Electr. Power Syst. Res. 2015, 120, 128–135. [CrossRef]
19. Swierczynski, M.; Stroe, D.-I.; Stan, A.-I.; Teodorescu, R.; Laerke, R.; Kjaer, P.C. Field tests experience from
1.6 MW/400 kWh Li-ion battery energy storage system providing primary frequency regulation service.
In Proceedings of the 2013 4th IEEE/PES Innovative Smart Grid Technologies Europe (ISGT EUROPE),
Lyngby, Denmark, 6–9 October 2013; pp. 1–5.
20. Stein, K.; Tun, M.; Matsuura, M.; Rocheleau, R. Characterization of a Fast Battery Energy Storage System for
Primary Frequency Response. Energies 2018. submitted.
21. Dubarry, M.; Devie, A. Battery durability and reliability under electric utility grid operations: Representative
usage aging and calendar aging. J. Energy Storage 2018, 18, 185–195. [CrossRef]
22. Dubarry, M.; Devie, A.; Stein, K.; Tun, M.; Matsuura, M.; Rocheleau, R. Battery Energy Storage System
battery durability and reliability under electric utility grid operations: Analysis of 3 years of real usage.
J. Power Sources 2017, 338, 65–73. [CrossRef]
23. Timothy, M. Spitler, “Lithium Ion Batteries”. U.S. Patent Application US20070092798A1, 23 October 2006.
24. HNEI. Development of Real-Time Closed-Loop Control Algorithms for Grid-Scale Battery Energy Storage Systems;
Hawaii Natural Energy Institute: Honolulu, HI, USA, 2014.
25. Kundur, P.; Balu, N.J.; Lauby, M.G. Power System Stability and Control; McGraw-Hill Eduation: New York, NY,
USA, 1994; Volume 7.
26. Existing Wind Farms. Available online: https://web.archive.org/web/20150222124557/http://
hawaiianelectric.com/heco/_hidden_Hidden/Renewable-Energy/Existing-Wind-Farms (accessed on
27 November 2018).

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).